In this paper, we propose a new framework for facial expression classification. This framework utilizes random forest as the classifier based on the features extracted from improved principal component analysis (PCA). Traditional PCA has two drawbacks: it is difficult to estimate the covariance matrix, and it is computational prohibitive to get the eigenvectors. In order to solve the two problems, we adopt an improved PCA to extract the features and uses the random forest algorithm as the classifier. The experiments are conducted based on the JAFFE facial expressions library, and are compared with the support vector machine (SVM) classifier. Theoretical analysis and experimental results show that the facial expression recognition system of random forest based on improved PCA extracting feature, whether in terms of facial expression recognition rate or running speed, has shown significant improvement.

The facial expression recognition method of random forest based on improved PCA extracting feature

Clustering validity index (CVI) plays an important role in data partitioning and image segmentation. In this paper, a new CVI is proposed to perform the color image segmentation. The proposed CVI combines compactness, separation and overlap to assess the clustering quality effectively. The aggregation operators (t-norms and t-conorms) are used to build a new reliable and robust overlap measure. Moreover, a genetic algorithm is employed to dynamically optimize the clusters centroids and get the best possible data partition. The clustering of super-pixels is performed to reduce the computational cost and convergence time. The genetic algorithm with new clustering validity index is able to find the best data partitioning. The performance of the proposed algorithm is evaluated on the Berkeley image segmentation database. The extensive experimentation shows that the proposed algorithm performs better compared to other state-of-the-art methods.

Color image segmentation using genetic algorithm with aggregation-based clustering validity index (CVI)

In this article we propose to combine an integrated method, the PCA-GMM method that generates a relatively improved segmentation outcome as compared to conventional GMM with Kalman Filtering (KF). The combined new method the PCA-GMM-KF attempts tracking multiple moving objects; the size and position of the objects along the sequence of their images in dynamic scenes. The obtained experimental results successfully illustrate the tracking of multiple moving objects based on this robust combination

/pdf/multiple-tracking-of-moving-objects-with-kalman-filtering-4tq706aigz.pdf

Multiple Tracking of Moving Objects with Kalman Filtering and PCA-GMM Method

Cette these est consacree a l'evaluation des familles exponentielles pour les problemes de la modelisation des bruits et de la segmentation des images couleurs. Dans un premier temps, nous avons developpe une nouvelle caracterisation des familles exponentielles naturelles infiniment divisible basee sur la fonction trace de la matrice de variance covariance associee. Au niveau application, cette nouvelle caracterisation a permis de detecter la nature de la loi d'un bruit additif associe a un signal ou a une image couleur. Dans un deuxieme temps, nous avons propose un nouveau modele statistique parametrique mulltivarie base sur la loi de Riesz. La loi de ce nouveau modele est appelee loi de la diagonale modifiee de Riesz. Ensuite, nous avons generalise ce modele au cas de melange fini de lois. Enfin, nous avons introduit un algorithme de segmentation statistique d'image ouleur, a travers l'integration de la methode des centres mobiles (K-means) au niveau de l'initialisation pour une meilleure definition des classes de l'image et l'algorithme EM pour l'estimation des differents parametres de chaque classe qui suit la loi de la diagonale modifiee de la loi de Riesz.

Contribution des familles exponentielles en traitement des images

This paper presents an unsupervised satellite color image segmentation approach based on Bivariate Beta type-II. Such a method could be considered as original since it uses a K-Means clustering algorithm in order to initialize the image class number. Moreover, it exploits a Bivariate Beta type-II for statistical distributions applying it for each class. Satellite image exploitation requires the use of different approaches, especially those based on the unsupervised statistical segmentation principle. Such approaches necessitate the definition of several parameters such as image class number, class variables estimation and mixture distributions. The use of statistical image attributes has allowed us to get convincing results, provided that we ensure under the condition of having an initialization step with appropriate statistical distributions. Bivariate Beta type-II associated with a K-means clustering algorithm and Expectation-Maximization (EM) algorithm could be adapted to such a problem. For each image class, Bivariate Beta type-II attributes a specific distribution type according to different parameters. Different adapted algorithms (namely K-Means clustering algorithm, EM algorithm and Bivariate Beta type-II algorithm) are then applied to the satellite image segmentation problem. The efficiency of those combined algorithms is validated with the Mean Squared Errors (MSE), Signal to Noise Ratio (SNR) and Maximum Distance (MD).

Unsupervised satellite image segmentation using a Bivariate Beta type-II mixture model

Image segmentation is one of the most important area of image retrieval. In colour image segmentation the feature vector of each image region is ’n’ dimension different from grey level image. In this paper a new image segmentation algorithm is developed and analyzed using the finite mixture of doubly truncated bivariate Gaussian distribution by integrating with the hierarchical clustering. The number of image regions in the whole image is determined using the hierarchical clustering algorithm. Assuming that a bivariate feature vector (consisting of Hue angle and Saturation) of each pixel in the image region follows a doubly truncated bivariate Gaussian distribution, the segmentation algorithm is developed. The model parameters are estimated using EM-Algorithm, the updated equations of EM-Algorithm for a finite mixture of doubly truncated Gaussian distribution are derived. A segmentation algorithm for colour images is proposed by using component maximum likelihood. The performance of the developed algorithm is evaluated by carrying out experimentation with six images taken form Berkeley image dataset and computing the image segmentation metrics like, Probabilistic Rand Index (PRI), Global Consistency Error (GCE), and Variation of Information (VOI). The experimentation results show that this algorithm outperforms the existing image segmentation algorithms.

/pdf/image-segmentation-method-based-on-finite-doubly-truncated-53qqkzch0a.pdf

Image Segmentation Method Based On Finite Doubly Truncated Bivariate Gaussian Mixture Model with Hierarchical Clustering

Colour Image segmentation is one of the prime requisites for computer vision and analysis. Much work has been reported in literature regarding colour image segmentation under HSI colour space and Gaussian mixture model (GMM). Since the Hue and Saturation values of the pixel in the image are non-negative. And may not be meso-kurtic, it is needed left truncate the Gaussian variate and is used to represent these two features of the colour image. The effect of truncation can not be ignored in developing the model based colour image segmentation. Hence in this paper a left truncated bivariate Gaussian mixture model is utilized to segment the colour image. The correlation between Hue and Saturation plays a predominant role in segmenting the colour images which is observed through experimental results. The expectation maximization (EM) algorithm is used for estimating model parameters. The number of image segments can be initialization of the model parameters are done with K-means algorithm. The performance of the proposed algorithm is studied by calculating the segmentation performance techniques like probabilistic rand index (PRI), global consistency error (GCE) and variation of information (VOI). The utility of the estimated joint probability density function of feature vector of the image is demonstrated through image retrievals. The image quality measures obtained for six images taken from Berkeley image dataset reveals that the proposed algorithm outperforms the existing algorithms in image segmentation and retrievals.

/pdf/studies-on-colour-image-segmentation-method-based-on-finite-22xro59vtq.pdf

Studies on Colour Image Segmentation Method Based on Finite Left Truncated Bivariate Gaussian Mixture Model with K-Means

Image segmentation plays a dominant role in image analysis and image retrievals. Much work has been reported in literature regarding image segmentation based on Gaussian mixture model (GMM). The main drawback of GMM is regarding the assumption that each image region is characterized by Gaussian component, in which the feature vector is mesokurtic and having infinite range. But in colour images the feature vector is represented by Hue and Saturation which are non- negative and may not be symmetrically distributed. Hence the image segmentation can not be accurate unless the non-negative nature of the feature vector is included. In this paper an image segmentation method is developed and analyzed with the assumption that the bivariate feature vector consisting of Hue and Saturation of each pixel follows a left truncated bivariate Gaussian mixture model. In this method the number of components (Image regions) are determined by Hierarchical clustering. The segmentation algorithm is proposed under Bayesian frame with maximum likelihood. The experimentation with six images taken from Berkeley dataset reveals that the proposed image segmentation method outperforms the existing image segmentation method with GMM and finite left truncated bivariate Gaussian mixture model with K-means.

G. V. S. Rajkumar

Papers

Image Segmentation Method Based On Finite Doubly Truncated Bivariate Gaussian Mixture Model with Hierarchical Clustering

Studies on Colour Image Segmentation Method Based on Finite Left Truncated Bivariate Gaussian Mixture Model with K-Means

Colour Image Segmentation with Integrated Left Truncated Bivariate Gaussian Mixture Model and Hierarchical Clustering